The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill attempts to read configuration preferences from a local file (CLAUDE.md) using grep and instructs the agent to use the found value without further verification.
Ingestion points: The CLAUDE.md file (Step 2) is read to determine the worktree directory path.
Boundary markers: None. The agent is instructed to use the preference "without asking."
Capability inventory: The skill possesses the ability to execute shell commands (git, mkdir, echo), write to files (.gitignore), and install software packages.
Sanitization: No sanitization or validation of the input retrieved from CLAUDE.md is specified before it is used in subsequent shell commands.
[COMMAND_EXECUTION]: The skill automatically executes build and installation scripts based on the presence of project files.
Evidence: The "Post-Creation Setup" section automatically triggers npm install, cargo build, pip install, poetry install, and npm test based on the existence of package.json, Cargo.toml, or requirements.txt.
Risk: This executes code (install scripts or test suites) defined within the repository being worked on.
[EXTERNAL_DOWNLOADS]: The skill triggers the retrieval of external dependencies from public registries.
Evidence: Commands like npm install, go mod download, and poetry install fetch packages from external sources (npm, PyPI, Go Proxy, etc.) during the worktree setup process.

using-git-worktrees